Post on 12-May-2022
Improved Hammond’s Landform Classification and Method for Global
250-m Elevation Data
Deniz Karagülle Environmental Systems Research Institute, Inc. Charlie Frye Environmental Systems Research Institute, Inc. Roger Sayre Land Change Science Program, U.S. Geological Survey Sean Breyer Environmental Systems Research Institute, Inc. Peter Aniello Environmental Systems Research Institute, Inc. Randy Vaughan Environmental Systems Research Institute, Inc. Dawn Wright Environmental Systems Research Institute, Inc.
Short Title: 250-m Global Hammond’s Landform Methodology
Key Words: Hammond, Landform, Classification, Global, Methodology, Land Surface
Form
Corresponding Author: Deniz Karagülle, Environmental Systems Research Institute,
Inc., 380 New York St, Redlands, CA 92373 USA. Email: dkaragulle@esri.com
2
3
Abstract
In 1964, E.H. Hammond proposed criteria for classifying and mapping physiographic regions of the United States. Hammond produced a map entitled “Classes of Land Surface Form in the Forty-Eight States, USA”, which is regarded as a pioneering and rigorous treatment of regional physiography. With the advent of computer processing and availability of digital elevation models (DEMs), several researchers have since developed digital landform models to map the distribution of physiographic regions using Hammond’s classification. These models have been applied in number of areas with varying size and physiographic complexity, and have emphasized different classification criteria and raster processing techniques. We propose a new algorithm for mapping Hammond’s regional landform classes using a global 250-meter DEM, and we characterize the results of that model with a new map of global Hammond landforms. The new algorithm incorporates slope, relative relief, and profile type as parameters, and uses varying neighborhood analysis sizes to characterize these parameters. We compare the new map to other global landform maps and compare its visual alignment with a global multi-directional hillshade dataset. We present potential applications of the map, and discuss important distinctions between physiographic regions and discrete landforms.
4
1 Introduction
Landforms are natural features on the Earth’s surface that both reflect and shape geophysical
and ecological process. The result is a defining part of landscapes that so often impact on
human perception and interactions with environment. Blascyznki (1997) defines landforms as,
“specific geomorphic features on the surface of the Earth, ranging from large-scale features
such as plains and mountain ranges to minor features such as individual hills and valleys,” and
suggests that the analysis and quantification of the Earth’s surface contributes to a better
understanding of the physical, chemical and biological progressions that appear within the
landscape. Landforms are one of the most important and intrinsic elements of landscape
analysis and exploration (Booth, 1983). Landforms provide a physical context for describing the
landscape, topography, and ecological units within the environment (MacMillan et al., 2000).
Understanding the physical and historical context of the landscape is necessary in order to
understand the temporal and spatial scales of ecosystems. Landforms are ecologically
important elements because ecosystems develop within landform regions, and material and
energy flows occur within the landform system (Swanson et al., 1988).
Landforms provide temporal and spatial perspectives for examining soil, vegetation, and
aquatic characteristics, and for interpreting ecosystem pattern and process (Bailey, 2006).
Landforms also mediate, bind, and influence macroclimates and microclimates. Bailey (2006)
writes:
On a mesoscale within the same macroclimate, we commonly find several broad-scale landform patterns that break up the zonal pattern. Hammond’s (1954) landform classification (based on surface geometry) is useful in capturing this effect on zonal climate and provides a basis for further differentiation of ecosystems, known as landscape mosaics.
Swanson et al. (1988) characterized four effects of landforms on ecosystem pattern and
process:
5
1. Elevation, steepness of slope, and aspect, in combination with temperature, moisture,
and nutrients within the landform region produce many different patterns in the physical
setting, and these patterns determine ecological potential.
2. Landform regions affect flow of organisms, propagules, energy and material.
3. Landform regions may affect the spatial pattern of non-geomorphic disturbance by fire
and wind.
4. Landforms constrain the spatial pattern, frequency, and rate of geomorphic processes
that alter biotic features and processes.
The understanding of landform distributions is therefore key to mapping and
understanding ecosystem patterns and distributions (Blasczynski, 1997). The importance of this
understanding has been recognized in ecosystem mapping approaches which incorporate
landform as a fundamental element in the modeling process (Sayre et al., 2014).
1.1 Hammond’s (1954) Landforms
Many geographers have described landforms from a genetic perspective, based on the
geomorphological processes that formed the landscape (Zakrzewska, 1967). Murphy (1968), for
example, mapped global physiographic regions from analysis of tectonic origins, macroclimate,
and intensity of dissection of the landscape by drainage channels. While genetic approaches
focus on delineating large area (e.g. continental and sub-continental scales) physiographic
regions, and emphasize the origins of terrain development, they are less suitable for
characterizing changes in relief and physiographic structure(Hammond, 1954).
Hammond (1954) identified challenges in applying genetic landform classifications to
small-scale (global / less detailed) representations of the landscape, citing problems of
inaccurate, overly complex, and sometimes ambiguous results. In 1964 he developed and
published a quantitative, empirical methodology for classifying and mapping landforms using a
set of morphological characteristics (Hammond, 1964). The approach has been considered a
6
macro-morphological classification based on the shape of the landform (Brabyn, 1998), where
individually mapped landscape features are associated with landform regions (Chorley, 1967).
Hammond (1954) emphasized that small scale characterization of landforms requires regional
characterization rather than individual feature identification:
In all small-scale maps, much is left out, notably the arrangement of features within the areas distinguished, but it is hoped that in general the most diagnostic and distinctive characteristics have been shown. Since the different terrain types are capable of quantitate definition, rough inter-regional comparison is possible. Further the representation is based on a system which permits any desired amount of expansion of detail as the scale of mapping increased. Finally, the maps do not attempt to indicate the probable origin of the terrain features but present the configuration empirically as an element of the natural landscape. (Hammond, 1954 pg. 42)
Hammond’s (1954) landform classification methodology was considered “systematic,
relatively objective, applicable to all types of terrain, suitable for medium and small scale
mapping, and useful for individual and comparative landform studies” (Zakrzewska, 1967).
Hammond’s (1954) extensive field experience provided the basis for the empirical definition of
slope, local relief, and profile type as the essential parameters for classifying and mapping
regional landform classes. Slope indicates the flatness or steepness of a location on the surface
of the Earth. Local relief is the amount of elevation change within an analysis area. Profile
separates uplands from lowlands, evaluates the amount of gentle slope within the uplands, and
identifies tablelands. Hammond assessed those parameters for the conterminous United States
through visual analysis of 1:250,000 topographic maps, and produced a pioneering map entitled
“Classes of land surface form in the forty-eight states, USA” (Hammond, 1964). This remarkable
resource was developed essentially in the field using topographic maps, prior to the availability
of advanced computer technologies and geographic information systems (GIS) (Smith and
Clark, 2005). At that time, the approach could not be implemented globally due to a general lack
of global topographic information and the immensity of the manual effort required for deriving
quantitative landforms without computers.
7
Today, with the availability of digital elevation models (DEMs) and GIS technology, it is
possible to map landforms using quantitative models and computer visualization (Smith and
Clark, 2005). GIS has become a fundamental and essential tool in the management and
analysis of quantitative data (Dikau et al.1991). In a series of innovative terrain
characterizations, Dikau and others (1991) became the first to automate Hammond’s semi-
quantitative system using GIS, producing a digital landform map of New Mexico (Dikau et al.,
1991 and 1995). They used the same parameters (slope, relief and profile) to classify landform
regions, but while Hammond used a 1:250,000 scale topographic map, Dikau et al. (1991) used
a 200-meter resolution DEM. Hammond used a 9.65 km neighborhood analysis window (NAW)
as the local analysis region within which the three parameters were computed. The Dikau et al.
(1991) analysis used a 9.8 km NAW. Subsequently, several researchers have followed a similar
GIS methodology in mapping landform classes for other specific study areas (Brabyn, 1998;
Gallant et al., 2005; Morgan and Lesh, 2005; Hrvatin and Perko, 2009).
True (2002) published a map of the Landforms of the Lower Mid-West (excerpted,
Figure 1) using an alternative method, which omitted the profile type parameter. For that
analysis, the 30-m spatial resolution National Elevation Dataset (NED) (Gesch et al., 2002) was
used with a 1-km NAW. This method, also referred to as the MoRAP (Missouri Resource
Assessment Partnership) (True, 2002) approach, is easier to implement as it uses only slope
and local relief. Although it can result in a granular texture, and regions can sometimes appear
as fragmented blends of two or more landform classes, the output is generally accurate.
The MoRAP method was subsequently implemented by the United States Geological
Survey (USGS) and others to map Hammond’s landforms of South America (Sayre et al., 2008),
the conterminous United States (Sayre et al., 2009), Africa (Sayre et al., 2013), and most
recently, the world (Sayre et al., 2014). The global landforms (excerpted, Figure 2) were
developed as inputs to a global ecosystem mapping effort (Sayre et al., 2014), with the
8
understanding that landforms are critical elements of ecosystem structure (Bailey, 2006). This
(Sayre et al., 2014) global landform map was developed from a 250 m global DEM ( Danielson
and Gesch, 2011) using the MoRAP approach and a 1 km NAW. This (Sayre et al., 2014)
global landform map was the first of its kind, produced at a relatively high spatial resolution (250
m) for the world.
9
Figure 1 Subset of the 30-m spatial resolution map ‘Landforms of the Lower Midwest (True, 2002) produced using the MoRAP (Missouri Resources Assessment Partnership) model. This portion of the map shows the Missouri River Valley between Iowa and Nebraska.
10
Figure 2 Subset of 250-m World Landforms-MORAP Method 2014 dataset. This portion of the map shows the Missouri River Valley between Iowa and Nebraska (Esri, 2014).
11
We used a panel of in house experts with experience in different areas of the world. We
reviewed areas these experts knew, by using a multidirectional hill shade. This review yielded
the following conclusions: First, landforms appeared to be strongly faceted and fragmented as
opposed to regionally homogenous. Second, some important terrain types (e.g. tablelands)
appeared to be missing. Finally, some areas classified as mostly plains were popularly regarded
as regions of hills or low hills; examples include the Wisconsin Dells, a highly incised terrain,
where only the slopes were characterized as hills; and the Flint Hills in eastern Kansas, which
are terraced hills. These discrepancies were revealed through spot check comparisons of
regions that had the words hills or escarpments in the name (e.g. Flint Hills), but the
corresponding locations were not so classified. A global, multi-directional hill-shaded terrain
dataset (Esri, 2014) was used for these comparisons, and the missing classes were found to be
globally consistent. There was an interest in improving the capture of “missing” relief that
occurred in large regions classified as flat or irregular plains. It was additionally noted that the
MoRAP method could not produce ecologically significant landform classes such as tablelands
because the MoRAP method did not take into account Hammond’s (1954) profile type
parameter.
As the review proceeded, it was felt that the “missing” relief may have been caused by
omitting profile type, the third of Hammond’s (1954) classification parameters, as a classifier. It
was also felt that the profile type parameter might be essential for imbuing a less-fragmented,
regional quality to the resulting map. Some of the fine granularity that was evident in the product
may have resulted from using a NAW that was too small, indicating that an important ratio exists
between the resolution of the DEM and the size of the analysis window. Moreover, it was felt
that varying the size of the NAW based on which of the three parameters being assessed
(slope, local relief, and profile type) might produce better results than using a fixed NAW size for
each of the three classifiers. Jasiewicz and Stepinski (2013) used a 30 meter DEM to map the
12
distinct terrain geometries (geomorphones) of Poland. They found that adjusting NAW size to
local relief improved the delineation of these landform elements, and so incorporated a varying
NAW into their model.
1.2 Goals and Objectives
There are many ways to identify landforms and landform regions. The overall goal of this study
is to identify regional landform classes phenomenologically adequate to characterize and define
ecosystems. This paper is primarily intended as a description of the GIS procedures used to
generate global Hammond landforms from 250 m DEM data. Our intention is to present these
techniques and procedures in sufficient detail as to be replicable by others at the level of
detailed GIS instructions. In addition to providing these procedures, our primary objectives for
this analysis were to 1) further test and characterize the relationships between classifier
parameters and NAW sizes, 2) evaluate a new model to improve the upon 2014 USGS/MoRAP
global landforms map, and 3) produce a new global Hammond landforms map. We evaluated
the use of an alternative model to the MoRAP algorithm seeking to improve regionalization in
the landform classification in order to update the map of global ecological land units produced
by Sayre et al., in 2014. We felt that an improved, better-regionalized landform map would
reduce the spatial complexity of the ecological land units (ELUs) initially produced from that
model.
2 Methodology
As an alternative to the MoRAP approach, we implemented a GIS-based articulation of
Hammond’s model (Morgan and Lesh, 2005) as an ArcGIS geoprocessing script tool, with the
script written in Python. The implementation used the same 250-m source DEM, the Global
Multi-resolution Terrain Elevation Data (GMTED) 2010 (USGS, 2010) used to develop the
original global landforms map (Sayre et al., 2014). Morgan and Lesh (2005) had successfully
13
followed Hammond’s (1954) landform classification approach based on the Dikau et al. (1995)
model using 30-meter resolution DEM data as their input source data to map landforms in the
state of Maryland. The Morgan and Lesh spatial model, developed for local landform mapping
using a 30-m input DEM, was used in a global application using 250-m resolution DEM. The
global application of the Morgan and Lesh (2005) model required identification of optimal NAW
sizes.
The Morgan and Lesh (2005) model used 36 steps to derive landforms from a DEM.
Their model first projects the DEM to a projected coordinate system suitable for distance-based
calculations, then produces classifications of slope, relief, and profile type before combining
them into a single expression of Hammond (1954) landform classes. Morgan and Lesh
implemented their model using a 30-m resolution DEM in the Maryland State Plane projected
coordinate system as their input data. They used a 600-m radius circular neighborhood (NAW)
to establish regional character. Their input DEM extent was for a 7.5 minute USGS quadrangle
representing approximately 55 square miles (142.4 square kilometers). The model produced
twenty-four Hammond landform classes. These landform classes are shown in Table 1.
14
Table 1 Morgan and Lesh (2005) Twenty-Four Landform Classes
Mountains Hills
High Mountains Low Mountains Open High Mountains Open Low Mountains Plains with Low Mountains Plains with High Mountains
Plains with Hills Plains with High Hills Open Very Low Hills Open Low Hills Open Moderate Hills Open High Hills Very Low Hills Low Hills Moderate Hills High Hills Tablelands Plains
Tablelands with Moderate Relief Tablelands with Considerable Relief Tablelands with High Relief Tablelands with Very High Relief
Flat or Nearly Flat Plains Smooth Plains with Some Local Relief Irregular Plains with Low Relief Irregular Plains with Moderate Relief
15
Instead of the Maryland State Plane projected coordinate system used by Morgan and
Lesh, this study projected the GMTED 2010 from the original WGS 1984 geographic coordinate
system to World Equidistant Cylindrical. This ensured that the same size of analysis window
could be used everywhere in the world. We projected the GMTED data using the ArcGIS Project
Raster geoprocessing tool, with the bilinear sampling option. There was no change in the
minimum or maximum data values and there was no significant change in the overall sum of
values.
As discussed above, the evaluation of the True (2002) and Sayre et al. (2014) landform
outputs had led us to conclude that the optimal size of a moving NAW for calculating slope,
relief and profile parameters is resolution dependent. To determine the optimal NAW size and
shape, we empirically tested several neighborhood sizes. Dikau et al. (1995) used a 200-meter
resolution DEM, and a 9.8 km NAW. Hammond (1954) used 1:250,000 scale topographic maps
instead of a DEM and a moving window of 9.65-kilometers. Therefore our 250 m global
implementation of the Morgan and Lesh model, we tested 12.5 km, 6 km, and 3 km NAW sizes.
2.1 Gentle Slope
We used ArcGIS to calculate percent slope using a 12.5 cell (3 km) NAW. Each cell in the NAW
is assigned a value that is the average of the slopes between that cell and the eight surrounding
cells, and these slope values are then averaged within the NAW. Hammond (1954) uses the
amount of gentle slope within the NAW to classify slope into four categories (Table 2). The
definition of gentle used was “any inclination of less than 8%” (Hammond, 1964 pg 15).
Appendix A contains a list of the steps used to calculate slope.
16
Table 2 Gentle slope classes and description.
Percent of Neighborhood Over 8% Slope Gentle Slope Class
0 - 20% 400
21 - 50% 300
51 - 80% 200
81 - 100% 100
Depending on relative local relief, DEM resolution affects derived slope values. To demonstrate
this, we evaluated three levels of local relief: high (greater than 900 meters), medium (between
150 meters - 300 meters), and low local relief (less than 90 meters) using four different DEM
resolutions. The results are shown in Table 3.
Table 3 Slope percentages by different resolutions
DEM Resolution High Relief Medium Relief Low Relief
250m 25.5% 4% 0.5%
90m 27% 4.5% 1.6%
30m 30.3% 7% 2%
10m 37% 10.5% 2.6%
Percent change from 10 meter
32% 61% 80%
The percentages in table 3 were produced using these steps. First we used local relief dataset
generated as part of this work. Next, we chose ten locations well within each relief zone and
recorded slope values for each different DEM resolution. We calculated mean values for these
sample locations. The change in slope is inversely proportionate to the change in DEM
resolution. This analysis of slope and DEM resolution validates the 8 percent threshold as
recommended by Hammond.
17
2.2 Calculation of Local Relief
Local relief is the amount of elevation change within a circular 6 km (25 cell) NAW. Hammond
(1954) defined six classes (Table 4) of local relief. Appendix B contains a list of steps used to
calculate Local Relief.
18
Table 4 Relief classes and description of how elevation changes in the 6 km analysis window.
Change in Elevation Relief Class
0 – 30 m 10
31 – 90 m 20
91 – 150 m 30
151 – 300 m 40
301 – 900 m 50
over 900 m 60
19
2.3 The Profile Parameter
The combination of Hammond’s (1954) slope and relief classes define some landform regions
like mountains, hills, or plains. However, the ecologically useful landform class of tablelands
requires the profile type parameter. The MoRAP approach used in the Sayre et al. (2014) global
landforms did not include tablelands as a landform region. Profile determines whether each cell
in the DEM is uplands or lowlands, and if uplands, the amount of gentle slope (Figure 3) is
calculated.
20
Figure 3 Diagrams of hypothetical slopes showing how each cell is evaluated and classified as either upland or lowland given the slope of other cells within a 6-km NAW (reproduced from similar diagrams in Dikau et al. (1995).
21
Tablelands only occur in uplands where at least 50% of the upland area within the NAW
contains gentle slopes. We calculated the profile parameter using a 25-cell (6-km) NAW (Figure
4).
22
Figure 4 Diagrams of hypothetical upland profiles showing Upland A, representing a tableland, Upland B representing mountains or hills depending on the slope and relief parameters. Lowlands represent plains and hills depending on the gentle slope and relief parameters.
23
Hammond (1954) defined four classes of profile type. The complexity of the profile type
parameter calculation required several iterations of development and testing. During these
iterations, we noted that earlier efforts to automate Hammond’s classification only accounted for
a portion of the potential range of output profile values. Therefore we produced five classes of
profile type (Table 5), introducing a class for 0.
24
Table 5 Profile classes and descriptions.
Percent of Neighborhood Over 8% Slope Profile Class
Under 50% of upland or lowland is gentle slope 0
Over 75% of lowland is gentle slope 1
50% - 75% of lowland is gentle slope 2
50% - 75% of upland is gentle slope 3
Over 75% of upland is gentle slope 4
25
The profile class of 0 (zero) represents upland areas of hills or mountains, and lowland
areas of hills. We do not interpret the omission for handling these cells in the earlier automation
efforts as anything more than a by-product of the complexity of the automation steps and
communicating them, because the affected areas were already implicitly classified correctly
using the slope and relief parameters. Appendix C lists the steps to compute profile classes.
2.4 Attribute Combination and Issues with Plains Classes
We calculated the global Hammond landform class by adding the slope, relief, and profile
outputs. We used the classifier combination schema from Morgan and Lesh (2005) where labels
contained three digit numbers. The hundreds digit identifies slope class, the ten digit identifies
local relief class, and the ones digit identifies profile type. Appendix D contains a master table of
combination outcomes for each attribute.
Our initial implementations revealed areas of plains within relief features that were
strongly influencing the landform class assignment. We visually evaluated the results with a
hillshade model overlay. Dual classes such as “Plains with Hills” or “Plains with Low Mountains”
appeared over-represented. For example, When the true landscape is Low Mountains, a
mountainous region was labeled “Plains with Low Mountains” instead of Low Mountains when
just a few plains cells occurred within the region. We interpreted these outcomes as the result of
using an inappropriate NAW size; specifically that plains features require a smaller NAW. For
this reason we extracted Plains features separately in a single global implementation to produce
plains vs. non-plains features. For that two-class implementation, we empirically determined to
use a 1 km NAW. We then “enforced” the plains features onto the multi-class global landform
model to allow all plains features to override the underlying classes. Appendix E contains the
steps used to calculate plains-only cells using slope and relief parameters with a 1-km
neighborhood.
26
2.5 The Final Landform Generation Model
Having now empirically evaluated appropriate NAW size to use for the classifiers, and devised
an adjustment to allow a plains-centric override approach, we finalized the global Hammond
landform modeling approach according to the schematic presented in Figure 5 below:
27
Figure 5 Conceptual Summary of the process to produce the Improved Hammond Landform Classification.
Input (250m DEM)
Gentle Slope
Profile
Local Relief
Combination of Classifiers
Feature-centric (Plains)
Adjustment
Improved Hammond Landform
Classification
28
2.6 Resampling
As is common in many raster-based classification and analyses, resampling of the modeled
output was necessary to remove or smooth the salt-and-pepper effect caused by a few isolated
cells within larger otherwise homogenous regions. The initial result of combining the slope, relief
and profile outputs included many one- and two- cell landform “features” within most regions.
For example, occurrences of isolated cells of hills or plains within regions of low or high
mountains were noted. These outlier classes made the larger surrounding regions appear
speckled and less definitive. We used the ArcGIS Spatial Analyst Focal Statistics tool with the
majority filter option and a 1-km circular window to aggregate these isolated cells into the larger
region. A “side-effect” of this tool is converting these cells to no data values when there is a tie
in the count of two or more types of majority cells within the NAW. Even though this
implementation used a larger neighborhood, the results did include some no data cells. These
cells were reset to their original value by reclassifying all No Data to 1000, then using the Local
Statistics tool, with the Minimum option. We then found that using the Majority Filter tool also
removed some of the artificial patterns from the circular NAW, and if run repetitively,
progressively addressed this problem. We successively filtered the data in this manner with four
iterations.
3 Results and Discussion
We first implemented the Morgan and Lesh (2005) model globally using a 12.5 km square NAW,
and although it processed relatively quickly, it produced numerous peculiar spatial artifacts
along the edges of landform classes. The results were visually evaluated using a hillshaded
terrain representation, and the authors felt that hilly and mountainous regions included too much
of the surrounding plains regions. Some regions contained ‘skirts’ of plains at the edges of
landform regions. We proceeded to implement the model with 6 km NAW size for local relief and
29
profile, and 3 km NAW size for the gentle slope, and evaluated the results. While the 6 km NAW
size produced good results for the local relief and profile parameters, a 3 km NAW generally
worked best for the slope parameter. These NAWs were empirically determined as the most
appropriate neighborhood analysis sizes for minimizing edge-related anomalies given the 250
meter DEM, and for best visual correlation with hillshade interpretations.
The final global landform analysis used a 3 km NAW for the slope classifier, and a 6 km
NAW for the local relief and profile type classifiers. In addition to identifying classifier-specific
NAW size for landform generation, we also determined appropriate NAW size (1 km) for a single
feature-specific (plains) implementation.
3.1 Plains-Centric Override Adjustment
The plains-centric adjustment produced a more intuitive result, without the problematic dual
landform classes like Plains with High Mountains. A graphic depiction of the result produced
from the plains override approach is presented in Figure 6. The plains override adjustment
eliminated (four) dual-landform classes.
30
Figure 6 Left image represents before overriding with Plain Areas. Right image represents after overriding with plain areas. Effect of applying a plains-centric override to remove edge-related classification anomalies. The hillshading provides a sense of where the terrain is most rugged. Note the removal of the artificial “skirt” of low mountains (left panel) which existed prior to the adjustment.
31
3.2 Data Filtering
The results of the application of the Focal Statistics Majority resampling process are
shown in Figure 7, below. Application of the filter removed numerous small inclusions and
contributed to a stronger sense of regionalization in the results. Visual inspection of the filtered
result against the hillshade dataset confirmed a match for numerous locations on all continents
for each class of landform.
32
Figure 7 Left panel represents before focal statistics majority, and right panel represents after focal statistics majority. Results from application of Focal Statistics Majority tool. Substantial filtering was achieved, using four iterations, leading to a smoother and regionalized appearance.
33
3.3 The Improved Global Hammond Landforms Product
The landforms produced from this analysis represent a substantial revision of the global
Hammond landforms produced using the MoRAP method and presented in Sayre et al. (2014).
The new landforms are depicted in Figure 8.
Figure 8 The improved map of global Hammond landform classification with sixteen landform classes.
These revised landforms demonstrate a clear transition from facet-style landform
characterization produced by the MoRAP approach (Sayre, et al. 2014) to a more intuitive
expression of regional landform classes. An iterative review process ensured that the resulting
classes were regional, intuitive, and suitable for contributing to the description of ecological land
units. The application of the improved Hammond’s algorithm to the GMTED 2010 250-meter
global DEM produced 16 landform classes (Table 6) compared to Morgan and Lesh’s (2005), 26
classes. There were two reasons for the reduction in classes: 1) removal of dual classes
containing plains, and 2) four classes from Dikau et al., (1991) were not realized with a 250-
meter DEM as opposed to a 30-meter DEM. Figure 9 presents a comparison of the original
Landform Classification Map (Sayre et al., 2014) with the new global Hammond landforms for
the Grand Canyon region in the United States.
36
Table 6 Classes produced by the Improved Hammond Landform Classification model
Mountains Hills
High Mountains Scattered Moderate Hills
Low Mountains Scattered High Hills
Scattered High Mountains Moderate Hills
Scattered Low Mountains High Hills
Tablelands Plains
Tablelands with Moderate Relief Flat or Nearly Flat Plains
Tablelands with Considerable Relief Smooth Plains with Some Local Relief
Tablelands with High Relief Irregular Plains with Low Relief
Tablelands with Very High Relief Irregular Plains with Moderate Relief
Figure 9 Comparison of the Improved Hammond’s Landform classes (the new map on the right side) produced using the Morgan and Lesh (2005) (right map) approach with the global landforms from Sayre et al. (left map) (2014), for a portion of the Grand Canyon in the U.S.
Two significant improvements are noted from the comparison of landforms in the Grand
Canyon shown in Figure 9. First, our implementation of the Morgan and Lesh (2005) model
produces a less fragmented, more regionalized characterization than the Sayre et al (2014)
implementation. Moreover, our model captures more hills landform regions in a plains matrix.
The improved physiographic regionalization and the improved capture of the plains with relief
features were the primary motivations for revising the Sayre et al., (2014) global landforms map.
Figure 10 is another comparison between the Sayre et al. (2014) and our
implementation of the Morgan and Lesh (2005) model, and includes the original Hammond
(1964) map. The three characterizations are presented for the State of New Mexico. These
comparisons show the progressive increase in terrain feature detail from Hammond (1964) to
Sayre et al. (2014) to this study. The original Hammond (1964) characterization is the most
regionalized, as would be expected from a semi-quantitative, manually constructed, non-digital
approach. The Hammond (1964) physiographic regions are represented as relatively large and
homogenous polygon regions, as opposed to the Sayre et al. (2014) approach in which
increased terrain feature granularity and heterogeneity is observed, but where the original sense
of general physiographic regionalization is lacking. The Morgan and Lesh (2005) model
provides a greater sense of regionalization, and the regions are generally visually consistent
with the original Hammond (1964) regions.
Figure 10 Left Map represents Hammond (1964), middle map represents USGS (2014) and right map represents Improved Hammond Landform classification map (this study). This is the comparison of results in New Mexico.
The improved global Hammond landforms dataset has sixteen classes, as opposed to
six landform classes modeled by Sayre et al. (2014). To facilitate a comparison of these two
resources, all classes for each dataset were aggregated into one of four classes: plains, hills,
low mountains, and high mountains (Table 7).
41
Table 7 Landform classes used Within the Four Major Landform Class Groups.
World Landforms (Sayre, et al. 2014)
Improved Hammond Global Landform Classification Map
Plains Flat or Nearly Flat Plains Smooth Plains with Some Local Relief
Flat or Nearly Flat Plains Smooth Plains with Some Local Relief
Hills High hills Moderate Hills Low Hills
High Hills Moderate Hills Scattered High Hills Scattered Moderate Hills Irregular Plains with Low Hills Irregular Plains with Moderate Relief Tablelands with Moderate Relief Tablelands with Considerable Relief Tablelands with High Relief Tablelands with very High Relief
Low Mountains
Low Mountains Low Mountains Scattered Low Mountains
High Mountains
High Mountains High Mountains Scattered High Mountains
42
Figure 11 shows the comparison of total area of each of the four generalized landform
regions (plains, hills, low mountains, and high mountains) for the two datasets. These results
are consistent with the assertion that the Sayre et al. (2014) product over-represented plains at
the expense of the other relief classes. The improved global Hammond landforms map identified
considerably less area in plains than the Sayre et al. (2014) map and a corresponding increase
in low and high mountains. The improved global Hammond landforms map therefore resulted in
a more balanced distribution of general landform classes, and reduced what had been noted by
the authors as an overabundance of plains features.
43
Figure 11 Comparison of major landform classes between the Sayre et al. (2014) map and Improved Hammond Landform Map (this study).
72%
21%
5% 0.1%
45%
19%
27%
9%
0
200,000,000
400,000,000
600,000,000
800,000,000
1,000,000,000
1,200,000,000
Plains Hills LowMountains
HighMountains
USGS 2014
Esri-Hammond's 2015
Sayre et al. (2014)
Improved Hammond
Landform Map (this
study)
44
While these four aggregated classes are useful for both comparing different landform
models and for aggregating ecologically meaningful areas by physiography, the improved
classification resolution in the new landform models is also welcome. The new model
characterizes 16 classes using a rigorous assessment of the three determinants: slope, local
relief, and profile type. The overall allocation of classifier attributes into final landform classes is
summarized in a master parameter combination diagram below (Figure 12). In this figure, all
landform classes are presented as composites of the values of the three parameters.
3.2 Final Landform Classes from Parameter Combinations
Figure 12 composites all three parameters to show how they relate to produce the final landform
classifications. There are two vertical axes on the diagram. The left shows local relief classes.
The right vertical axis shows profile classes within each relief class. The horizontal axis
indicates slope classes. The upper right portion of the table describes occurrences of higher
slopes combined with higher relief, resulting in high and low mountains. Hills result when the
relief is lower (150-300m) but the slope is still high. The lower left area in the table shows
situations of low slope and low relief, yielding the plains classes. The upper left portion
represents tablelands, occurring when the profile parameter is high (3-4). This figure also shows
the impact of the plains feature override, which split dual classes such as plains with low
mountains into separate regions for plains and low mountains. The plains features added in the
later stages of the process affected the upper left portion of the diagram and corresponded with
tablelands, where the profile parameter values were 0, 1, and 2.
Figure 12 Diagram of how landform classes are produced from the slope, local relief, and profile parameters, and how the additional plains features impacted these combinations.
4 Conclusions and Future Work
This study implemented the latest GIS-based articulation of Hammond’s model (Morgan and
Lesh, 2005) as an ArcGIS geoprocessing script tool, with the script written in the Python
programming language. This implementation used the Global Multi-resolution Terrain Elevation
Data (GMTED) 2010, a 250-meter resolution DEM. To our knowledge, this is the first successful
automated implementation of Hammond’s complete landform classification on a global DEM.
The result shows that this method produces landform classes consistent with the earlier work of
Hammond (1964) and Dikau et al.,(1995).
The regional landform classes produced in this work describe landscape morphology in
a fashion sufficient to support characterization and definition of ecosystems and ecological land
units. Additionally, the work to refine the neighborhood window sizes, and to subsequently
override the dual class regions with plains, addressed one of Dikau et al.’s (1995) suggestions
for future work. That suggestion questioned the value of regions with dual classes.
Compared to Hammond’s (1964) work, which produced a small scale U.S. map of
landforms, and Dikau et al.’s (1995) work produced a similar scale map of New Mexico, this
work produced more detailed results. This is due, at least in part, to the use of satellite derived
DEMs instead of DEMs based on topographic maps. The increased detail produces many
smaller regions of tablelands, hills, and mountains. Thus, one suggestion for future work is to
develop a set of minimum mapping units for each type of regional landform class, which might
reduce the micro-level detail of the product and further increase regionalization.
This work is an improvement over previous Hammond-related landform regionalization
for two reasons. First, and primarily, the elimination of the dual classes reduces the overall
number of classes, making the result easier to understand and to incorporate into other GIS
analyses and models. Second, separately calculating plains with a smaller neighborhood
47
analysis window and overriding the initial result with these plains produced results that match
expectations when comparing the landforms data with a hillshade.
Due to the relatively complicated series of steps required to calculate the profile type
parameter, it was necessary to investigate the output of each step to understand what was
produced. Doing so at several study area locations (Grand Canyon Region) provided the
necessary insight to identify tablelands. One insight gained was that escarpment features are
located at the edges of some tablelands. For this work, we considered escarpments as features,
rather than regional landform classes, and thus, did not pursue classifying them. We determined
that escarpments may be found by calculating slope and relief in a 1-km neighborhood window
and then looking at very steep and very high relief areas within tablelands or mountains.
Given the availability of global 30-meter DEMs, potential future work could include a
higher spatial resolution implementation of the model, which might identify additional features
such as terraces, escarpments, badlands, and canyons. A higher spatial resolution input DEM,
however, would likely change the approach to an identification of microscale, local landform
features (cliffs, toeslopes, ridges, etc) rather than mesoscale physiographic regions. The
classification of regions versus features does not, however, need to be mutually exclusive.
Harris et al. (2014) mapped the geomorphology of the seafloor by first characterizing four
general physiographic regions, and then overlaying on these regions discrete landform features
such as seamounts and submarine canyons. Such an approach could be adopted for the land
masses of the planet as well, mapping mesoscale Hammond landform regions first, and then
overlaying local terrain features onto the Hammond regional landform complexes.
Our work was framed in an empirical investigation of global landform distributions, and
we found that identifying the correct algorithm, parameter choice, and neighborhood window
size required careful experimentation and could vary among landform features to be classified.
Although in fidelity to Hammond’s classes we did not incorporate other landforms like terraces,
48
badlands, and canyons into our classification, we experimented with these classes and
concluded that they could be feasibly identified in future work. For example, identification of
canyons, which exhibit relative relief differences similar to mountains and hills, might be easily
separated from these classes by inverting the DEM.
Our work demonstrates that global Hammond (1964) landform classes can be
successfully modeled using a 250 m global DEM. Now that a finer spatial resolution global DEM
(30m) is available it might be tempting to try to implement the model with the finer resolution
data. Prior to such an effort, however, and especially given the immensity of the processing
effort that would be required to model global landforms at a 30 m resolution, the overall goal
would need to be clarified. If the goal is only to continue identifying the same types of landform
regions, then there may be no added value in producing spatially refined boundaries between
landforms regions, since these boundaries are imprecise to begin with, and related to human
interpretations of terrain variation. However, if there is an interest in a more detailed delineation
of landform features in regions, including the identification of additional feature classes, then
pursuing a global landforms implementation with finer source data may be warranted.
Finally, while the methodological development has been presented herein, the
application value or intended utility of the new global landform resource needs to be
established. We are using the resource as an input to the characterization of ecologically
distinct areas. Other “non-ecological” applications are envisioned as well, ranging from
assessments of agricultural productivity potential to assessments of movement of disease
vectors across landscapes. We welcome feedback from the global ecosystems community and
other interested communities on the utility of the resource.
49
4 Availability
Improved Hammond Landform Classification Map and underlying data can be accessible via
ArcGIS Online and Living Atlas via
http://www.arcgis.com/home/item.html?id=cd817a746aa7437cbd72a6d39cdb4559
Acknowledgements
The comments and suggestions of Michael Dangermond, Dr. Mark Kumler, Dean Gesch, Bruce
Worstell greatly improved this manuscript.
List of Figures
Figure 1 Subset of the 30-m spatial resolution map ‘Landforms of the Lower Midwest (True,
2002) produced using the MoRAP (Missouri Resources Assessment Partnership) model. This
portion of the map shows the Missouri River Valley between Iowa and Nebraska.
Figure 2 Subset of 250-m World Landforms-MORAP Method 2014 dataset. This map shows
Bakersfield, CA (Esri, 2014).
Figure 3 Diagrams of hypothetical slopes showing how each cell is evaluated and classified as
either upland or lowland given the slope of other cells within a 6-km NAW (reproduced from
similar diagrams in Dikau et al. (1995).
Figure 4 Diagrams of hypothetical upland profiles showing Upland A, representing a tableland,
Upland B representing mountains or hills depending on the slope and relief parameters.
Figure 5 Conceptual Summary of the process to produce the Improved Hammond Landform
Classification.
50
Figure 6 Effect of applying a plains-centric override to remove edge-related classification
anomalies. The hillshading provides a sense of where the terrain is most rugged. Note the
removal of the artificial “skirt” of low mountains (left panel) which existed prior to the adjustment.
Figure 7 Results from application of Focal Statistics Majority tool. Substantial filtering was
achieved, using four iterations, leading to a smoother and regionalized appearance.
Figure 8 The improved map of global Hammond landform classification with sixteen landform
classes.
Figure 9 Comparison of the Improved Hammond’s Landform classes produced using the
Morgan and Lesh (2005) approach with the global landforms from Sayre et al. (2014), for a
portion of the Grand Canyon in the U.S.
Figure 10 Hammond (1964) USGS (2014) and Improved Hammond Landform classification map
(this study) comparison of results in New Mexico.
Figure 11 Comparison of major landform classes between the USGS (2014) map and Improved
Hammond Landform Map (this study).
Figure 12 Diagram of how landform classes are produced from the slope, local relief, and profile
parameters, and how the additional plains features impacted these combinations.
51
References
Bailey G R 2006 Research Application of Ecosystem Patterns. Proceedings of the Eight Annual Forest Inventory and Analysis Symposium: 83-90
Barka I, Vladovic J, and Malis F 2011 Landform classification and its application in predictive mapping of soil and forest units. GIS-Ostrava: 23-26
Blaszczynski S J 1997 Landform Characterizations with Geographic Information Systems. Photogrammetric Engineering & Remote Sensing 63:183-19
Booth K N 1983 Basic Elements of Landscape Architectural Design. Long Grove IL, Waveland Press
Brabyn L 1996 Landscape classification using GIS and national digital databases. Landscape Research. London, Taylor and Francis
Brabyn L 1998 GIS analysis of macro landform. In P. Firns (Ed.), SIRC98 – The 10th Annual Colloquium of the Spatial Information Research Centre. Dunedin, University of Otego: 35–48
Chorley R J and Hagget P 1967 Physical and Information models in Geography. Norcross, GA: The Trinity Press
Cress J J, Sayre R, Patrick P, and Warner H 2009 Terrestrial Ecosystems—Land Surface Forms of the Conterminous United States, Scientific Investigations Map 3085 Version 1.0
Danielson J and Gesch D 2011 Global Multi-resolution Terrain Elevation Data 2010 (GMTED 2010). U.S Geological Survey Open File Report 2011-1073. 26p
Dikau R, Brabb E E, and Mark R M.1991 Landform classifications of New Mexico by computer. U.S. Geological Survey: 91-634
Dikau R, Brabb E E, Mark R K, and Pike R J 1995 Morphometric landform analysis of New Mexico. Zeitschrift für Geomorphologie supplement and 101: 109-12
Gallant L A, Douglas B D, and Hoffer R M 2005 Automated Mapping of Hammond’s Landforms. IEEE Geoscience and Remote Sensing Letters 2, No. 4: 384-388
Gesch D, Oimoen M, Greenlee S, Nelson C, Steuck M, and Tyler D 2002 The National
Elevation Dataset, American Society For Photogrammetry and Remote Sensing 68 No 1
Harris P T, Macmillan-Lawler M, Rupp J, and Baker E K 2014 Geomorphology of the Oceans. Marine Geology 352: 4-24
Hammond H E 1954 Small-scale continental landform maps. Annals of the Association of American Geographers 44 No 1: 33-42
Hammond H E 1964 Classes of land surface form in the forty-eight States, USA. Annals of the Association of American Geographers 54, no. 1, map sup. No. 4, 1:5,000,000
Hrvatin M, and Perko D 2009 Suitability of Hammond’s Method for determining landform units in Slovania. Acta geographica Slovenica 49: 343-366
52
Jasiewicz J, and Stepinski T F 2013 Geomorphons-a pattern recognition approach to classification and mapping of landforms. Geomorphology 182: 147-156.
MacMillan R A, McNabb D H, and Jones R K 2000 Automated landform classification using DEM’s: a conceptual framework for a multi-level, hierarchy of hydrologically and geomorphologicaly oriented physiographic mapping units. In 4th International Conference on Integrating GIS and Environmental Modeling: 1-18.
Morgan J, and Lesh A 2005 Developing landform maps using ESRI’S Model Builder. Presented at ESRI International User Conference’05.
Murphy R E 1968 Landforms of the world. Annals Map, Supplement 9. Annals, Association of American Geographers 58.
Pike R J, and Rozema W J 1975 Spectral Analysis Of Landforms. Annals of the Association of American Geographers65: 499–516.
Sayre, Roger, Bow, Jacquie, Josse, Carmen, Sotomayor, Leonardo, and Touval, Jerry, 2008. Terrestrial ecosystems of South America, chap. 9 of Campbell, J.C., Jones, K.B., Smith, J.H., and Koeppe, M.T., eds., North America Land Cover Summit: Washington, D.C., Association of American Geographers, p. 131–152. (Also available online at http://www.aag.org/books/nalcs/.) Sayre, Roger, Comer, Patrick, Warner, Harumi, and Cress, Jill, 2009. A new map of standardized terrestrial ecosystems of the conterminous United States: U.S. Geological Survey Professional Paper 1768, 17 p. (Also available online at http://pubs.usgs.gov/pp/1768
Sayre, R., P. Comer, J. Hak, C. Josse, J. Bow, H. Warner, M. Larwanou, E. Kelbessa, T. Bekele, H. Kehl, R. Amena, R. Andriamasimanana, T. Ba, L. Benson, T. Boucher, M. Brown, J. Cress, O. Dassering, B. Friesen, F. Gachathi, S. Houcine, M. Keita, E. Khamala, D. Marangu, F. Mokua, B. Morou,
L. Mucina, S. Mugisha, E. Mwavu, M. Rutherford, P. Sanou, S. Syampungani, B. Tomor, A. Vall, J. Vande Weghe, E. Wangui, and L. Waruingi. 2013. A New Map of Standardized Terrestrial Ecosystems
of Africa. Washington, DC: Association of American Geographers. 24 pages.
Sayre R, Dangermond J, Frye C, Vaughan R, Aniello P, Breyer S, Cribbs D, Hopkins D, Nauman R, Derrenbacher W, Wright D, Brown C, Convis C, Smith J, Benson L, Paco VanSistine D, Warner H, Cress J, Danielson J, Hamann S, Cecere T, Reddy A, Burton D, Grosse A, True D, Metzger M, Hartmann J, Moosdorf N, Dürr H, Paganini M, DeFourny P, Arino O, Maynard S, Anderson M., and Comer P 2014 A New Map of Global Ecological Land Units — An Ecophysiographic Stratification Approach. Washington, DC: Association of American Geographers.
Swanson F J, Kratz T K, Caine N, and Woodmansee R G 1988 Landform Effects on Ecosystem Patterns and Processes. BioScience 38 No. 2.
Smith J M, and Clark C D 2005 Methods for visualization of digital elevation models for landform mapping. Earth surface process landforms 30: 885-900.
Stepinski T F 2011 A new approach to classification of landforms. Proceedings of Geomorphometry. Geomorphology: 109-112.
Zakrzewska B 1967 Trends and Methods in Land Form Geography. Annals of the Association of American Geographers 57, No.1: 128-165.
WWW Documents: Esri 2014 World Landforms– MORAP Method 2014. WWW document, http://www.arcgis.com/home/item.html?id=c8bf5dbce9214f7d8e17d3179e6e802e.
53
WWW Documents: Esri 2014 Terrain: Multi-Directional Hillshade. WWW document, http://landscapeteam.maps.arcgis.com/home/item.html?id=53924d5afc2a4edb805a72c64710b6b3
WWW Documents: True D. 2002. Landforms of the Lower Mid-West, Missouri Resource Assessment Partnership (MoRAP): Columbia, Mo., University of Missouri MoRAP Map Series MS–2003–001, scale 1:1,500,000. WWW Documents: http://www.cerc.usgs.gov/morap/Maps.aspx?MapID=7 .
WWW Documents: USGS (2010). Global Multi-resolution Terrain Elevation Data, United States Geological Survey. WWW document, http://topotools.cr.usgs.gov/gmted_viewer/.
54
Appendix A
Gentle Slope Calculations
1. Use the ArcGIS Desktop Spatial Analyst Extension’s Slope tool to produce the percent
slope with the GMTED 2010 DEM as the input data. The range of slopes produced was
0% to 742.269%.
2. Use the ArcGIS Desktop Spatial Analyst Extension’s Reclassify tool with the percent
slope dataset from step 1 such that values less than 8% are set to 0, and values greater
than 8% slope are set to 1.
3. To calculate percent of gentle slope in the 3km neighborhood window, use the ArcGIS
Desktop Spatial Analyst Extension’s Focal Statistics tool. The input dataset is the output
of step 2, and the neighborhood type is circular. Use the Sum option, which assigns the
sum of the cells within the neighborhood to each cell in the output.
4. Reclassify the result of step 3 is the four categories defined by Hammond (1964) (Table
2).
55
Appendix B
The steps to assign local relief to each cell in the GMTED 2010 DEM are:
1. Use the ArcGIS Spatial Analyst Extension’s Focal Statistics tool to calculate maximum
elevation within 25-cell circular neighborhoods with the GMTED 2010 DEM as the input
data.
2. Use the ArcGIS Spatial Analyst Extension’s Focal Statistics tool to calculate minimum
elevation within 25-cell circular neighborhoods with the GMTED 2010 DEM as the input
data.
3. Use the ArcGIS Spatial Analyst Extension’s Minus tool to subtract the output of step 1
from the output of step 2.
4. Reclassify the result of step 3 into the six categories defined by Hammond (Table 3).
56
Appendix C
The steps to assign profile to each cell in the GMTED 2010 DEM are:
1. Use ArcGIS Spatial Analyst Math tools to calculate the half of the local relief. To do so,
divide the output of difference of maximum and minimum elevation, by 2. This
represents the local point of distinction between lowlands and uplands.
2. Use the ArcGIS Spatial Analyst Focal Statistics tool with the Minimum option and a 25-
cell size. Add this to the output of step 1. This represents the elevation of the local
distinction between lowlands and uplands.
3. Use the ArcGIS Spatial Analyst Minus tool to subtract the output of step 2 from DEM.
Because the DEM values are higher, this produces a surface where the positive values
represent lowlands, and negative values represent uplands.
4. Use ArcGIS Spatial Analyst Reclassify tool to reclassify lowlands as 1, and uplands as 2.
5. Use ArcGIS Spatial Analyst Reclassify tool to set lowland areas to 1 and uplands to 0.
6. Identify gentle slopes within lowlands by multiplying the output of step 5 by the output of
step 2 in slope parameter.
7. Use the output of step 6 as input to the ArcGIS Spatial Analyst Focal Statistics tool with
the Sum to calculate the sum of gentle slopes in lowlands.
8. Use Map Algebra, Float tool to create a floating point of step 3 in slope parameter.
9. Divide the output of Step 7 by the output of Step 8 to create a percentage of gentle
slopes in lowlands. Note: after the first iteration, we saw that Step 9 produced a result
with No Data values. This was resolved by inserting steps 10 and 11.
10. Use ArcGIS Spatial Analyst Reclassify tool to reclassify the output of step 9 NoData
values to 0 (zero); all other values are greater than zero and less than or equal to one.
11. With the output of Steps 9 and 10 as inputs, use ArcGIS Spatial Analyst Cell Statistics
with the Minimum option.
57
12. Multiply the output of step 11 by the output of Step 5 to isolate gentle slopes in lowlands.
13. Use ArcGIS Spatial Analyst Focal Statistics with output of Step 12, a 25-cell circular
neighborhood, and the Sum option to calculate the sum of gentle slopes.
14. Use ArcGIS Spatial Analyst Reclassify tool with the output of step 13 to reclassify
percentage of gentle slopes in lowland areas (<50% = 0, 50%-75% = 2, >75% = 1) Note:
Unlike the approaches of Hammond, Dikau et al., and Morgan and Lesh,we broke the
percent area occupied by lowlands into three classes instead of two. Previous models
did not account for areas under 50% of gentle slope in lowland areas.
15. Use ArcGIS Spatial Analyst Reclassify tool with the output of Step 4 to reclassify
uplands as 1 and lowlands as 0. (1=0, 2=1)
16. Identify gentle slopes by multiplying the output of step 15 with step 2 in slope parameter.
17. Use the output of Step 16 as input to the ArcGIS Spatial Analyst Focal Statistics tool with
the Sum to calculate the sum of gentle slopes in uplands.
18. Calculate the percentage of gentle slopes in uplands by dividing the output of Step 17 by
the output of Step 8. Note: Similar to Step 9, we saw that Step 17 also produced a result
with No Data values. This was resolved by inserting steps 19 and 20.
19. Use ArcGIS Spatial Analyst Reclassify tool to reclassify the output of step 18 to set No
Data values to 0 (zero); and all other values to one.
20. With the output of Steps 18 and 19 as inputs, use ArcGIS Spatial Analyst Cell Statistics
with the Minimum option.
21. Multiply the output of step 20 by the output of Step 15 to isolate gentle slopes in uplands.
22. Use ArcGIS Spatial Analyst Reclassify tool to reclassify the output of Step 21 to produce
the percentage of gentle slopes in upland areas (<50% = 0, 50%-75%= 3, >75%=4)
23. Add the output of Steps 22 and 14 to get the combination of percentage of gentle slopes
within lowlands and uplands.
58
Appendix D
Landform Combination and Classification Schema used by Morgan and Lesh (2005).
Combined Values Morgan and Lesh (2005) Classification
Slope (hundreds)
Relief (tens)
Profile (ones)
140-141-142-143-144 Low Mountains 100 40 0-1-2-3-4
150-151-152-153-154 Low Mountains 100 50 0-1-2-3-4
160-161-162-163-164 High Mountains 100 60 0-1-2-3-4
230 Moderate Hills 200 30 0
231-232-233-234 Open Moderate Hills 200 30 1-2-3-4
240 High Hills 200 40 0
241-242-243-244 Open High Hills 200 40 1-2-3-4
250 Low Mountains 200 50 0
251-252-253-254 Open Low Mountains 200 50 1-2-3-4
260 High Mountains 200 60 0
261-262-263-264 Open High Mountains 200 60 1-2-3-4
320 Irregular Plains with Low Hills 300 20 0
321-322-323-324 Irregular Plains with Moderate Relief 300 20 1-2-3-4
330 Irregular Plains with Hills 300 30 0
331-332 Plains with Hills 300 30 1-2
333-334 Tablelands with Moderate Relief 300 30 3-4
340 Irregular Plains with High Hills 300 40 0
341-342 Plains with High Hills 300 40 1-2
343-344 Tablelands with Considerable Relief 300 40 3-4
350 Irregular Plains with Foothills 300 50 0
351-352 Plains with Low Mountains 300 50 1-2
353-354 Tablelands with High Relief 300 50 3-4
360 Irregular Plains with Spires 300 60 0
361-362 Plains with High Mountains 300 60 1-2
363-364 Tablelands with Very High Relief 300 60 3-4
410 Nearly Flat Plains 400 10 0
411-412-413-414 Flat or Nearly Flat Plains 400 10 1-2-3-4
420 Nearly Flat Plains 400 20 0
421-422-423-424 Smooth Plains with some local relief 400 20 1-2-3-4
430 Nearly Flat Plains with Hills 400 30 0
431-432 Plains with Hills 400 30 1-2
433-434 Tablelands with Moderate Relief 400 30 3-4
440 Nearly Flat Plains with High Hills 400 40 0
441-442 Plains with High Hills 400 40 1-2
443-444 Tablelands with Considerable Relief 400 40 3-4
450 Nearly Flat Plains with Low Mountains 400 50 0
451-452 Plains with Low Mountains 400 50 1-2
453-454 Tablelands with High Relief 400 50 3-4
460 Nearly Flat Plains with High Mountains 400 60 0
461-462 Plains with High Mountains 400 60 1-2
463-464 Tablelands with Very High Relief 400 60 3-4
59
Appendix E
The steps to calculate feature centric plains are:
1. Use the ArcGIS Desktop Spatial Analyst Extension’s Slope tool to produce the percent
slope with the GMTED 2010 DEM as the input data. The range of slopes produce was
0% to 742.269%.
2. Use the ArcGIS Desktop Spatial Analyst Extension’s Reclassify tool with the percent
slope dataset from Step 1 such that values less than 5% are set to 0, and values greater
than 5% slope are set to 1.
3. Use ArcGIS Majority Filter with 8 neighborhoods use the input as the output of Step 2
4. Use ArcGIS Desktop Spatial Analyst Extension’s Focal Statistics tool with the Sum
option and a circular neighborhood of 4 cells (1-km), and the output of Step 3 as the
input.
5. Reclassify the result of Step 4 into two categories. 0 to 20% as 1 and over 20% as No
Data. Note: Slope of the neighborhood window is considered flat when more than 80%
of the area in a 1-km neighborhood has slope less than 5 percent of slope, otherwise the
area is considered as not flat area therefore as classified as No Data, eliminating those
cells from the next steps.
6. Use the ArcGIS Spatial Analyst Extension’s Focal Statistics tool to calculate maximum
elevation within 4-cell circular neighborhood using the GMTED 2010 DEM as the input
data.
7. Use the ArcGIS Spatial Analyst Extension’s Focal Statistics tool to calculate minimum
elevation within 4-cell circular neighborhoods with the GMTED 2010 DEM as the input
data.
8. Use the ArcGIS Spatial Analyst Extension’s Minus tool to subtract the output of step 7
from the output of Step 6
60
9. Reclassify the result of Step 8 into the two categories. 0-30-ms as 10 and rest is No
Data. Note: Local relief is classified as very low when the difference between maximum
and minimum elevation is between the ranges of 0-30-ms, otherwise the area is
considered as not low relief and classified as No Data.
10. Use the ArcGIS Spatial Analyst Extension’s Plus tool to add the output of step 9 to the
output of step 5. The combination of slope and local relief class determines the feature
centric plain areas.
11. Use the ArcGIS Spatial Analyst Extension’s Reclassify tool to reclassify only plain areas
to 2000. The input for this step is the output of 10.
12. Use the ArcGIS Spatial Analyst Extension’s Cell Statistics maximum tool with the output
of step 11 and the previous final map to override the plains features.
13. Use the ArcGIS Spatial Analyst Extension’s Reclassify tool, to reclassify the output of
map 12, select unique values with the Table 6. Green row indicates as the changing
values. This step removes the dual-classes.
14. Use ArcGIS Spatial Analyst Extension, Cell statistics maximum tool to add surface water
to the output of map 13.